Running Stockfish on a Google Cloud Virtual Machine

[bvdahl]

Disclaimer: I've also posted this question on Google Groups (without much help so far)

------------------------------------------

Hello all,

I am very much a noob at this, but recently I had the idea to run Stockfish on a very high-specced Google Cloud VM.

I managed to set up a 96 Core (Intel Xeon 2GHz) machine with 360 GB of RAM. Installed ChessBase, and set up Stockfish as a UCI engine, and thought I would be ready to crush the speed records.

The only default setting I changed was to use 96 threads.

What happened was that first of all, it took a long time to load the engine (about 3 minutes until it started calculating lines). Once it did load, however, it ran fine.

The minor disappointment was that it ran at about 65,000 kN/s, while the best Cloud engine that I can access runs at about 100M kN/s or about 50% more. This cloud engine runs 40 Cores Dual Xeon E5-2679 v4 Processors at 3.3 GHz.

Is the difference down to the difference in frequency of the processors of the two machines, or can I make adjustments in the settings / setup to make Stockfish run faster. Maybe it's not such a bright idea to try this out on Google Cloud after all? Also, it doesn't seem to matter which 64 bit version of SF I use.

Any help, or pointers to where I can read some relevant material would be greatly appreciated.

Best regards,
Baard Dahl

[Joost Buijs]

Disclaimer: I've also posted this question on Google Groups (without much help so far)

------------------------------------------

Hello all,

I am very much a noob at this, but recently I had the idea to run Stockfish on a very high-specced Google Cloud VM.

I managed to set up a 96 Core (Intel Xeon 2GHz) machine with 360 GB of RAM. Installed ChessBase, and set up Stockfish as a UCI engine, and thought I would be ready to crush the speed records.

The only default setting I changed was to use 96 threads.

What happened was that first of all, it took a long time to load the engine (about 3 minutes until it started calculating lines). Once it did load, however, it ran fine.

The minor disappointment was that it ran at about 65,000 kN/s, while the best Cloud engine that I can access runs at about 100M kN/s or about 50% more. This cloud engine runs 40 Cores Dual Xeon E5-2679 v4 Processors at 3.3 GHz.

Is the difference down to the difference in frequency of the processors of the two machines, or can I make adjustments in the settings / setup to make Stockfish run faster. Maybe it's not such a bright idea to try this out on Google Cloud after all? Also, it doesn't seem to matter which 64 bit version of SF I use.

Any help, or pointers to where I can read some relevant material would be greatly appreciated.

Best regards,
Baard Dahl

A Google cloud VM is implemented as a hyper-thread, instead of 96 real cores you probably got 48 cores with hyper-threading, that explains the bad results. 2 GHz. CF is also a bit low.